ThinkSystem NVIDIA L40S 48GB PCIe Gen4 Passive GPU

The ThinkSystem NVIDIA L40S 48GB PCIe Gen4 Passive GPU based on the Ada Lovelace architecture, is a powerful universal GPU for the data center, delivering breakthrough multi-workload acceleration for large language model (LLM) inference and training, graphics, and video applications. As the premier platform for multi-modal generative AI, the L40S GPU provides end-to-end acceleration for inference, training, graphics, and video workflows to power the next generation of AI-enabled audio, speech, 2D, video, and 3D applications.

The following figure shows the ThinkSystem NVIDIA L40S 48GB PCIe Gen4 Passive GPU.

Figure 1. ThinkSystem NVIDIA L40S 48GB PCIe Gen4 Passive GPU

AI models are exploding in complexity and popularity with the disruption led by large language models (LLMs) such as ChatGPT and generative AI diffusion models. L40S’s fourth-generation Tensor Cores with the Transformer Engine and new FP8 data format enable AI performance that exceeds the NVIDIA A100 Tensor Core GPUs for many AI training and inference workloads.

The following table shows the ordering information for the NVIDIA L40S GPU.

The option part number includes the following:

• One NVIDIA L40S GPU with full-height (3U) adapter bracket attached
• Documentation

Generative AI is fueling transformative change, unlocking a new frontier of opportunities for enterprises across every industry. To transform with AI, enterprises need more compute resources, greater scale, and a broad set of capabilities to meet the demands of an ever-increasing set of diverse and complex workloads.

The NVIDIA L40S GPU is the most powerful universal GPU for the data center, delivering end-to-end acceleration for the next generation of AI-enabled applications—from generative AI and model training and inference to 3D graphics, rendering, and video applications.

Enterprises are looking to use mainstream infrastructure to satisfy their compute needs, but training state-of-the-art models requires massive compute capability. For LLM models, eight L40S’s in mainstream servers bring up to 1.7X the training performance of an NVIDIA HGX™ A100 8-GPU system, giving enterprises fast time to solution with traditional infrastructure. When compared to the A100 80GB SXM for inference, the L40S delivers up to 1.2X more generative AI inference performance using StableDiffusion and up to 1.5X inference performance on popular networks, such as those included within the MLPerf benchmark.

Key use cases of the NVIDIA L40S GPU:

• Generative AI

  The AI, graphics, and media acceleration capabilities of the L40S GPU make it the premier platform for multi-modal generative AI pipelines. With powerful inferencing capabilities, combined with NVIDIA RTX™-accelerated ray tracing and dedicated encode and decode engines, the L40S accelerates AI-enabled audio, speech, 2D, video, and 3D generative AI applications.

  For image generative AI inference, the L40S GPU delivers more than 5X higher performance than the previous-generation NVIDIA A40 GPU and 1.2X more performance than the HGX A100. This breakthrough performance, combined with 48GB of memory capacity, makes the L40S GPU the ideal generative AI platform for high-quality images and immersive visual content.

• LLM Inference and Training

  Accelerate training, fine tuning, and inference workloads with powerful throughput and floating-point performance to build and deploy state-of-the-art AI models. Powerful NVIDIA-Certified Systems™ with eight L40 GPUs can train foundational models with up to 175 billion parameters to convergence and accelerate fine-tuning and retraining of existing large-scale models to adapt them for new tasks.

  Combining NVIDIA’s full stack of inference serving software with the compute capabilities of the L40S provides a powerful platform for trained models ready for inference. With support for structural sparsity and a broad range of precisions, including TF32, INT8, and FP8, the L40S delivers over 1 petaFLOPS of inference operation performance, delivering actionable insights with speed and precision.

• AI-Ready Development Platform with NVIDIA AI Enterprise

  Enterprise adoption of AI is now mainstream and leading to an increased demand for skilled AI developers and data scientists. Organizations require a flexible, high-performance platform consisting of optimized hardware and software to maximize productivity and accelerate AI development.

  NVIDIA AI Enterprise is an end-to-end, enterprise-grade AI software platform that offers 100+ frameworks, pretrained models, and libraries to streamline development and deployment of production AI, including generative AI, computer vision, and speech AI. Optimized and certified for reliable performance, NVIDIA AI Enterprise, together with the L40S, provides a unified platform to develop applications once and deploy anywhere, reducing the risks involved with moving from pilot to production.

• Rendering and 3D Graphics

  Running professional 3D visualization applications with NVIDIA L40S enables creative professionals to iterate more, render faster, and unlock tremendous performance advantages that increase productivity and speed up project completion. The NVIDIA L40S's third-generation RT Cores and industry-leading 48GB of GDDR6 memory deliver up to 2X the real-time ray-tracing performance of the previous generation.

  With these capabilities, artists and designers can work with complex geometry and high-resolution textures in real time to generate photorealistic designs and power full-fidelity creative workflows, from interactive rendering to virtual production.

• NVIDIA Omniverse

  NVIDIA Omniverse is a multi-GPU-enabled open platform for Universal Scene Description (USD)-based collaboration and real-time photorealistic simulation. The full-stack platform based on USD and NVIDIA RTX is the powerful culmination of NVIDIA's core graphics, compute, and AI technologies. NVIDIA L40S GPUs bring powerful AI and RTX capabilities to accelerate 3D content creation and industrial digitalization.

  For the most complex Omniverse workloads like extended reality (XR), multi-user design collaboration, and digital twins, the NVIDIA L40S enables ray-traced and path-traced rendering of materials, physically accurate simulations, and generation of photorealistic 3D synthetic data.

• Streaming and Video Content

  The NVIDIA L40S takes streaming and video content workloads to the next level, delivering breakthrough media acceleration capabilities with three video encode and three video decode engines. With the addition of AV1 encoding, the L40S delivers up to 2X the performance and improved TCO for broadcast streaming, video production, and transcription workloads.

• Virtual Workstations

  When combined with NVIDIA RTX Virtual Workstation (vWS) software, the NVIDIA L40S can be virtualized to deliver high-performance workstation instances to remote users for high-end design, AI, and compute workloads. With 48GB of GPU memory, the NVIDIA L40S with vWS enables flexible, work-from-anywhere solutions for GPU memory-intensive workloads.

The NVIDIA L40S GPU has the following specifications:

• Form factor
  • PCIe Full Height Full Length adapter (4.4-in x 10.5-in), Double-width (dual slot)
  • NVIDIA Form Factor 5.5
• Host interface:
  • PCIe 4.0 x16
  • MSI-X interrupt messaging protocol (MSI not supported)
  • PCIe Lane Polarity Inversion and Lane Reversal
• Single Root I/O Virtualization (SR-IOV) support
  • 256 virtual functions (VFs)
  • ARI Forwarding
• Hardware Root of Trust
  • Secure boot
  • Secure firmware upgrade
  • Firmware rollback protection
  • Support for in-band firmware update disable (established after each GPU reset)
  • Secure application processor recovery

The following table lists the GPU processing specifications and performance of the NVIDIA L40S GPU.

The following tables list the ThinkSystem servers that are compatible.

The following table lists the supported operating systems.

This section lists the NVIDIA software that is available from Lenovo.

• NVIDIA vGPU Software (vApps, vPC, RTX vWS, and vCS)
• NVIDIA Omniverse Software (OVE)
• NVIDIA AI Enterprise Software
• NVIDIA HPC Compiler Software

• Virtual PC (vPC)

  This product is ideal for users who want a virtual desktop but need great user experience leveraging PC Windows® applications, browsers and high-definition video. NVIDIA Virtual PC delivers a native experience to users in a virtual environment, allowing them to run all their PC applications at full performance.

• NVIDIA RTX Virtual Workstation (RTX vWS)

  NVIDIA RTX vWS is the only virtual workstation that supports NVIDIA RTX technology, bringing advanced features like ray tracing, AI-denoising, and Deep Learning Super Sampling (DLSS) to a virtual environment. Supporting the latest generation of NVIDIA GPUs unlocks the best performance possible, so designers and engineers can create their best work faster. IT can virtualize any application from the data center with an experience that is indistinguishable from a physical workstation — enabling workstation performance from any device.

• Virtual Compute Server (vCS)

  NVIDIA Virtual Compute Server (vCS) enables data centers running on Red Hat Enterprise Linux, Red Hat Virtualization, and other supported KVM-based hypervisors to accelerate server virtualization with the latest NVIDIA data center GPUs, so that the most compute-intensive workloads, such as artificial intelligence, deep learning, and data science, can be run in a virtual machine (VM) powered by NVIDIA vGPU technology.

The following license types are offered:

• Perpetual license

  A non-expiring, permanent software license that can be used on a perpetual basis without the need to renew. Each Lenovo part number includes a fixed number of years of Support, Upgrade and Maintenance (SUMS).

• Annual subscription

  A software license that is active for a fixed period as defined by the terms of the subscription license, typically yearly. The subscription includes Support, Upgrade and Maintenance (SUMS) for the duration of the license term.

• Concurrent User (CCU)

  A method of counting licenses based on active user VMs. If the VM is active and the NVIDIA GRID vGPU software is running, then this counts as one CCU. A vGPU CCU is independent of the connection to the VM.

The following table lists the ordering part numbers and feature codes.

NVIDIA Omniverse Software (OVE)

Lenovo offers the following NVIDIA Omniverse Software software. 3D workflows are now an essential component of every industry. Efficient design collaboration between teams is crucial. NVIDIA Omniverse™ Enterprise is an end-to-end collaboration and simulation platform that fundamentally transforms complex design workflows, creating a more harmonious environment for creative teams.

• NVIDIA Omniverse Enterprise Nucleus

  Collaboration service which enables a variety of Omniverse Enterprise-enabled client applications (Apps, Connectors, and others) to share and modify authoritative representations of virtual worlds in USD. Omniverse Nucleus is built for collaboration across multiple users. Omniverse Nucleus is the database and collaboration engine of Omniverse.

  Includes:

  • Collaboration and scalable core microservices deployed on servers and/or workstations.
  • Nucleus Workstation
  • Enterprise Nucleus Server
  • NVIDIA Enterprise Support

  Note: The Omniverse Enterprise Nucleus Subscription is priced and licensed per named user.

• NVIDIA Omniverse Enterprise Creator

  App for accelerated advanced scene composition and world building. Creator enables scene composition, simulation, and rendering on workstations and virtual workstations.

  Includes:

  • Omniverse Create
  • Omniverse Kit
  • Omniverse Extensions
  • Batch microservices (on up to 64 GPUs)
  • NVIDIA Enterprise Support

  Note: Collaboration via Nucleus requires a Nucleus subscription.

• NVIDIA Omniverse Enterprise Reviewer

  Simple app designed to visualize projects in stunning, physically accurate photorealism and enable minor editing and comment permissions. Reviewer enables review and approval for workstations and virtual workstations.

  Includes:

  • Omniverse View
  • NVIDIA Enterprise Support

  Note: Editing and commenting via Nucleus requires a Nucleus subscription.

The following license types are offered:

• Concurrent User (CCU) License

  A method of allocating licenses based on the number of users that are concurrently using the software. As an example, one CCU license allows only one concurrent user to use the software. 

• Floating User License

  For licensing, same as Concurrent User License.

• Named User License

  A method of licensing based on the number of named authorized users who may not re-assign or share the license with any other person.

• Subscription

  A software license that is active for a fixed period as defined by the terms of the subscription. An annual subscription includes SUMS for the duration of the license term.

The following table lists the ordering part numbers and feature codes.

NVIDIA AI Enterprise Software

Lenovo offers the NVIDIA AI Enterprise (NVAIE) cloud-native enterprise software. NVIDIA AI Enterprise is an end-to-end, cloud-native suite of  AI and data analytics software, optimized, certified, and supported by NVIDIA to run on VMware vSphere and bare-metal  with NVIDIA-Certified  Systems™.  It includes key enabling technologies from NVIDIA for rapid deployment, management, and scaling of AI workloads in the modern hybrid cloud.

NVIDIA AI Enterprise is licensed on a per-GPU basis. NVIDIA AI Enterprise products can be purchased as either a perpetual license with support services, or as an annual or multi-year subscription.

The features of NVIDIA AI Enterprise Software are listed in the following table.

Note: Maximum 10 concurrent VMs per product license

The following table lists the ordering part numbers and feature codes.

Find more information in the NVIDIA AI Enterprise Sizing Guide.

NVIDIA HPC Compiler Software

The GPU option part number does not ship with auxiliary power cables. Cables are server-specific due to length requirements and the connector on the server end of the cable. For CTO orders, auxiliary power cables are derived by the configurator. For field upgrades, cables will need to be ordered separately as listed in the table below.

Table 12. Auxiliary power cables for NVIDIA L40S GPU

Auxiliary power cable needed with the SR675 V3

235mm 16-pin (2x6+4) cable Option: 4X97A84510, ThinkSystem SR675 V3 Supplemental Power Cable for H100 GPU Option Feature: BSD2 SBB: SBB7A65299 Base: SC17B39301 FRU: 03LE554

The NVIDIA L40S GPU has the following regulatory approvals:

• RCM
• BSMI
• CE
• FCC
• ICES
• KCC
• cUL, UL
• VCCI

The NVIDIA L40S GPU has the following operating characteristics:

• Ambient temperature
  • Operational: 0°C to 50°C (-5°C to 55°C for short term*)
  • Storage: -40°C to 75°C
• Relative humidity:
  • Operational: 5 to 85% (5 to 93% short term*)
  • Storage: 5 to 95%

* A period not more than 96 hours consecutive, not to exceed 15 days per year.

One year limited warranty. When installed in a Lenovo server, the GPU assumes the server’s base warranty and any warranty upgrades.

The following sales training courses are offered for employees and partners (login required). Courses are listed in date order.

1. NVIDIA L40S GPU Overview and Business Use Case
   2023-10-12 | 60 minutes | Employees Only
   Details
   NVIDIA L40S GPU Overview and Business Use Case
   
   

   Welcome to the NVIDIA L40S GPU Overview and Business Use Case course. This course offers a closer look at the L40S GPU, featuring a webinar presented by Brad Davidson from NVIDIA. Throughout this course, we delve deep into the L40S GPU's capabilities, provide situational use cases, guide you on effectively positioning the L40S in various scenarios, and facilitate a meaningful comparison between the L40S and DGX systems.
   
   Completing this course will enable you to:
   • Describe the basics of NVIDIA L40S
   • Discuss how NVIDIA L40S delivers level performance for AI
   • Discuss generative AI and omniverse

   Published: 2023-10-12
   Length: 60 minutes
   Employee link: Grow@Lenovo
   Course code: DAINVD102

For more information, refer to these documents:

• ThinkSystem and ThinkAgile GPU Summary:
  https://lenovopress.lenovo.com/lp0768-thinksystem-thinkagile-gpu-summary
• ServerProven compatibility:
  http://www.lenovo.com/us/en/serverproven
• Lenovo Reference Architecture for Generative AI Based on Large Language Models (LLMs)
  https://lenovopress.lenovo.com/lp1798-reference-architecture-for-generative-ai-based-on-large-language-models
• NVIDIA L40S product page:
  https://www.nvidia.com/en-us/data-center/l40s/